DeepSeek has launched its new feature, 'Deep Think', which allows users to observe the reasoning process of its language model. This mode, described as having visible thought traces, has garnered attention for its unique approach to problem-solving. Users have reported mixed results, with some noting that while DeepSeek's reasoning is engaging and educational, it struggles with certain challenges, such as the LOOP wordgrid evaluation. The DeepSeek-R1-Lite-Preview variant reportedly thinks for over six minutes, outperforming other models like GPT-4o and Claude 3.5 Sonnet in some tests. However, it has also faced criticism, failing to solve riddles and explain its reasoning in certain cases. Overall, the launch has sparked interest in the capabilities and limitations of this new AI model.